Implementation of Data Mining and Spatial Mapping in Determining National Food Security Clusterization

Abstract

This study proposes a cluster analysis of provinces based on national food security data. The research objective is to determine provincial clusters based on food indicators which include rice harvest area, distribution of rice stocks, percentage of trade margin and transportation of rice distribution, percentage of average per capita expenditure, and total per capita consumption of rice. The source of observation data for the Rice Harvested Area by Province variable is the Ministry of Agriculture, Central Bureau of Statistics and Agriculture Services throughout Indonesia. This study uses data mining techniques in data processing with the K-Medoids algorithm. The K-Medoids method is a clustering method that functions to break down data sets into several groups. The advantage of this method is that it can overcome the weakness of the K-Means method which is sensitive to outliers. Another advantage of this algorithm is that the results of the clustering process do not depend on the order in which the dataset is entered. The k-medoids clustering method can be applied to food security data by province. From grouping the data obtained three clusters, with silhouette coefficient values for cluster 1, cluster 2, and cluster 3 respectively 0.33; 0.32; and 0.44. With the largest silhouette coefficient value obtained in cluster 3 and the cluster has entered into a strong cluster structure. The research results can provide information to the government about food security grouping data in Indonesia which has an impact on the distribution and availability of food in Indonesia.

Keywords

• k medoids algorithm
• data mining
• food security data
• silhouette coefficient.